Winboard Forum

The source code for Glaurung 0.2.4 is now available for download. This version is not very thoroughly tested, but I think it is a bit stronger than 0.2.3. It should also have a more active and interesting style of play. Binaries and an updated version of Scatha will follow within a few days.

The most important change is that Glaurung finally has a mobility evaluation. I am not yet sure how much this improves playing strength (if at all), but it certainly makes the games look more attractive. Other changes worth mentioning are an improved pawn storm evaluation for positions with castling to opposite sides, a few new pieces of endgame knowledge, and numerous small search and evaluation tweaks.

Glaurung 0.2.4 does not use buggy futility pruning. I want to have a better understanding of why BFP works and replace it with something better and more thought out before I include it in an official version.

The UCI strength options are removed in this version. They never really worked well, and will be replaced with something better (not based on slowing down the engine) in some future version.

Tord

Another piece of news worth mentioning is that the lazy author has finally written a Readme.txt file for Glaurung.

Tord

Intel 'profile-guided' optimized build:

http://homepages.tesco.net/henry.ablett/glaurung024.rar

Jim.

Jim Ablett wrote:Intel 'profile-guided' optimized build:

http://homepages.tesco.net/henry.ablett/glaurung024.rar

Jim.

Thanks, Jim!

Meanwhile, my first results with the new version are rather depressing. Unless it looks a bit better when more games are played, I should find a new hobby.

Tord

Hi Tord,

I don't know how I have downloaded Glaurung 0.2.3, I will look again for 0.2.4. The version I intended to test did a crash under Arena:

Regards, Reinhard.

[Event "Chess960 20 min + 5 sec"]
[Site "CHESSBOX"]
[Date "2005.07.02"]
[Time "23:33:30"]
[Round "1"]
[White "Glaurung 0.2.3 (Arena)"]
[Black "Smirf Beta BC-070 (Smirf GUI)"]
[Result "0-1"]
[Annotator "R. Scharnagl"]
[SetUp "1"]
[FEN "nbrnbqkr/pppppppp/8/8/8/8/PPPPPPPP/NBRNBQKR w KQkq - 0 1"]

{Glaurung 0.2.3 crashed in Arena, thus the game has been terminated.} 1. c4 Nb6
{(12.00) -0.174} 2. c5 Nd5 {(13.18) -0.070} 3. Nb3 f5 {(12.21) +0.007} 0-1

Hi Tord,

with 0.2.4 the castling seems not to work correctly ...

Regards, Reinhard.

[Event "Chess960 20 min + 5 sec"]
[Site "CHESSBOX"]
[Date "2005.07.02"]
[Time "23:48:52"]
[Round "2"]
[White "Glaurung 0.2.4 (Arena GUI)"]
[Black "Smirf Beta BC-070 (Smirf GUI)"]
[Result "0-1"]
[Annotator "R. Scharnagl"]
[SetUp "1"]
[FEN "bnrnqkrb/pppppppp/8/8/8/8/PPPPPPPP/BNRNQKRB w KQkq - 0 1"]

{Game aborted, because after black's castling white started calculating with
the black pieces ???} 1. g3 d6 {(11.05+) -0.168} 2. Ne3 g6 {(12.00) -0.138} 3.
Nc3 Ne6 {(11.12) +0.053} 4. b3 c6 {(11.34) +0.331} 5. Ng2 O-O {(10.03) +0.595}
0-1

Tord Romstad wrote:Meanwhile, my first results with the new version are rather depressing. Unless it looks a bit better when more games are played, I should find a new hobby.

Tord, let's face it... it's now obvious that computer chess is not for you, so you should start looking for another hobby. Take arcade emulation for example, it's fresh and new and there's a lot of fun to be had. By a fortunate coincidence I happen to have an open source emulator of mine hanging over here, and considering you've been such a nice guy in the past I am now willing to exchange it for that old patzer of yours. I know I know... don't thank me. So, do we have a deal?

Tord Romstad wrote:The source code for Glaurung 0.2.4 is now available for download. This version is not very thoroughly tested, but I think it is a bit stronger than 0.2.3. It should also have a more active and interesting style of play. Binaries and an updated version of Scatha will follow within a few days.

The most important change is that Glaurung finally has a mobility evaluation. I am not yet sure how much this improves playing strength (if at all), but it certainly makes the games look more attractive.
Tord

I disagree
Here is some code for mobility in 0.2.3.
you change the code but old glaurung already had mobility for bishops


/* Compute mobility */
x = 0;
for(ptr = Directions[BISHOP]; *ptr; ptr++) {
for(tosq=sq+(*ptr); Board[tosq]==EMPTY; tosq+=(*ptr)) x++;
if(COLOUR(Board[tosq])==(side^1)) x++;
}
mobility[side]+=BMobBonus[x];
if(PawnRank[side][sq]==RANK_1) {
d++;
if(x<=2) d++;
}

Reinhard Scharnagl wrote:Hi Tord,

I don't know how I have downloaded Glaurung 0.2.3, I will look again for 0.2.4. The version I intended to test did a crash under Arena:

Hi Reinhard,

There seems to be a misunderstanding. It looks like you are trying to play FRC against Glaurung, but Glaurung is not FRC compatible. For FRC, download Gothmog (my old engine). You'll find it here.

Tord

Alessandro Scotti wrote:

Tord Romstad wrote:Meanwhile, my first results with the new version are rather depressing. Unless it looks a bit better when more games are played, I should find a new hobby.

Tord, let's face it... it's now obvious that computer chess is not for you, so you should start looking for another hobby. Take arcade emulation for example, it's fresh and new and there's a lot of fun to be had. By a fortunate coincidence I happen to have an open source emulator of mine hanging over here, and considering you've been such a nice guy in the past I am now willing to exchange it for that old patzer of yours. I know I know... don't thank me. So, do we have a deal?

Arcade emulation? Why not?

I just have to see a few more results before I accept your deal.

Tord

Uri Blass wrote:

Tord Romstad wrote:The most important change is that Glaurung finally has a mobility evaluation. I am not yet sure how much this improves playing strength (if at all), but it certainly makes the games look more attractive.
Tord

I disagree
Here is some code for mobility in 0.2.3.
you change the code but old glaurung already had mobility for bishops

You're right, all previous versions of Glaurung had a simple mobility calculation for bishops, but nothing at all for the other pieces. I don't think this is quite enough to say that Glaurung had a real mobility eval.

I am sorry for the ugly loop unrolling in the mobility calculations in 0.2.4. The code turned out to be too slow with all the loops.

Tord

Hi Tord,

Tord Romstad wrote:

Reinhard Scharnagl wrote:I don't know how I have downloaded Glaurung 0.2.3, I will look again for 0.2.4. The version I intended to test did a crash under Arena:

There seems to be a misunderstanding. It looks like you are trying to play FRC against Glaurung, but Glaurung is not FRC compatible. For FRC, download Gothmog (my old engine). You'll find it here.

then I am very sorry about that. I errornously had assumpted, that Arena would refuse to play Chess960 with unaware engines. I should have had a more intensive look into Glaurung's and Arena's docs.

Regards, Reinhard.

I did a small test of 0.2.4 v BFP, at 4'+2", 128M hash each, ponder off, amd 3000xp+, same book, and this is the result I got:

1 Glaurung BFP 12.0/20
2 Glaurung 0.2.4 8.0/20

Of course this is not conclusive of anything.

Hi Tord,
I was pondering the BFP question yesterday and I wonder if the following test could be of help. Replace FP with:
1) set a prune flag according to (standard) FP;
2) set a prone flag according to BFP;
3) do not prune but search the tree normally;
4) if FP and BFP do not agree, see which one is right and save the position.

I know that not pruning at step 3) is going to change the tree shape with respect to when pruning is enabled, but maybe the test will still provide some insight.

Alessandro Scotti wrote:Hi Tord,
I was pondering the BFP question yesterday and I wonder if the following test could be of help. Replace FP with:
1) set a prune flag according to (standard) FP;
2) set a prone flag according to BFP;
3) do not prune but search the tree normally;
4) if FP and BFP do not agree, see which one is right and save the position.

Hi Alessandro,

I did many such experiments in Gothmog. As expected, BFP made many more mistakes than FP. That BFP works at all means that although it makes a lot of mistakes, these mistakes mostly happen in positions which are not important. The inaccuracies in BFP are biggest when the absolute value of the eval is huge, and these positions are usually already won or lost anyway.

I am not at all sure that ugly, accidentally discovered BFP I am using is very good. However, I think my experiments with BFP indicate that traditional futility pruning is a very non-optimal technique, and that it would be possible to replace it with something similar but much more effective. I'll let you all know if I find something, of course.

Tord